Elon Musk: Tesla Doesn’t Need Lidar, Will Tweak Radar To Be Lidar-Like


Yesterday, we posted on ZF’s CEO stating that autonomous driving requires Lidar, in addition to Radar and a bunch of cameras. The article itself suggested that Tesla disagrees with this lidar requirement. Here’s why ZF’s CEO, Stefan Sommer, believes lidar is necessary:

โ€œFor autonomous driving, we will need three core technologies: picture processing camera technology, short and long-range radar and lidar.โ€

โ€œLaser or infrared-based lidar technology will help vehicle sensors pick up contours and contrasts of obstacles which normal cameras are unable to detect, particularly in low light situations.โ€

Well, immediately following publication of our article, Tesla CEO Elon Musk took to Twitter to further strengthen our suggestion that Tesla is against lidar.

Musk posted the following on Twitter:

Elon Musk Tweets

Elon Musk Tweets

In general, those against lidar take that stance due to the technology currently being cost prohibitive for mass-produced automobiles.

However, it seems Elon Musk is rather confident that radar can be modified in such a way as to closely mimic the capabilities of lidar.

Musk Tweet

Musk Tweet

Perhaps this is the only way to improve autopilot without a substantial increase in cost to end consumers.

Guess we’ll find out soon if Tesla is able to achieve what Musk thinks is doable.

Categories: Tesla

Tags: ,

Leave a Reply

40 Comments on "Elon Musk: Tesla Doesn’t Need Lidar, Will Tweak Radar To Be Lidar-Like"

newest oldest most voted

I was waiting for the Master plan 2! It is Friday already, was he not supposed to tell us about it this week?

You missed the top-secret part. It probably has been unveiled to those that need to know ๐Ÿ˜‰

It was a double secret unveiling. You know, like in “Animal House” where Delta House was put on double secret probation.

Musk is right, sort of. Radar resolution is a function of aperture and bandwidth. With synthetic aperture of phased array _and_ wide bandwidth, it will outperform lidar. But I don’t think that’s what Tesla’s using. Still, human beings don’t have radar nor lidar, and we seem to do “ok”.

Except in heavy fog/rain/snow. Many people just pull over because they can’t see (or creep at 5mph). Others don’t. Then you get speed differentials.. which are what kill.

How much would a human brain, eyes, and ears cost to install in a vehicle?

Isn’t that called an Uber, or just mom for many? ๐Ÿ˜‰

If autopilot takes off, it will be cheap due to the supply side surge.

We are nowhere near matching the human brain. We just have lots of idiots in our industry who think we are.

Respectfully, maybe you want to take some time to update your knowledge…

You mean like Watson being Jeopardy champ, Deep blue beating chess champs, AlphaGo beating go champ? Tech is moving, and I’m hoping full auto will be ready by the time I get Tesla 3. I will be napping or reading Insideevs in my commute.

Even something as simple as autopilot when in traffic and audible warning to wake me up when traffic clears would be huge help. Traffic isn’t so bad when you can use that time to do something other than mindless watching of the car in front.

Deep Blue has NOTHING to do with AI – it’s specific algorithms for chess playing and has no other applications (there’s also the strong suggestion of cheating, but that’s another matter).

AI is scary stuff mainly because of the runaway process that would happen very quickly indeed if at some point it becomes marginally smarter than humans. A brain implemented in silicon easily has the potential to be a million times faster than our slow neurons, and the AI could therefore design a better version of itself very fast, then reiterate, and hours later the AI will already be incomparably smarter than any human…

I find it ironic that the perhaps two greatest threats to our species survival are artificial intelligence and natural stupidity!

So do you think this superAI is going to make us pets in a zoo? It reminds me of the Star Trek episode “The Menagerie” where Kirk would rather die on his feet, than live on his knees as a pet in a fake world.

I hope we don’t become like Neo in The Matrix, where you have to choose between fighting the machines and living in the real world, or playing Pokemon in a fake world created by AI. I would side with Neo. Do you?

I think we are close to matching the brains of those idiots though. So ironically, in their own limited worldview, they are correct. (LOL)

The idea that lidar is necessary is silly, since human’s have only ever used a dual camera system for driving, and with much lower resolutions than digital camera’s have now. It just turns out that the human brain is really really good at image processing.

Eventually image processing will improve to the point where you won’t need radar, lidar, sonar, or any other “ar’s” to have self driving cars.

So true – one of those ‘why didn’t I think of that’ insights that no one thinks of (at least that I have ready anywhere about it.

“Eventually image processing will improve to the point where you wonโ€™t need radar, lidar, sonar, or any other ‘arโ€™s’ to have self driving cars.”

Eventually, yes, altho relatively safe driving in all conditions will require use of infrared frequencies which can see thru fog, something the human eye is not capable of.

But the computer processing power, and the extremely sophisticated software needed for reliable optical shape/object recognition, at the moment seem distant goals. Progress in this field has been very slow. Using today’s technology, either radar or lidar appear to be the only practical technologies for self-driving cars’ sensors.

Sure. Give it another 100 years or so.

To quote someone in hurly-burly: not a ducking thing is necessary!

Seatbelts and airbags were unnecessary too once upon a time. French president Mitterand used to say we must not bother the people with seat belts and speed limits.

But progress happens. If infrared light or lidar or other incredibly useful and sexy tech makes for safer and/or faster autonomous cars, it’s obviously interesting to use it. How well humans drive is IRRELEVANT!

Besides, humans are still far superior to any computer in image processing. It’s not a given this will change, certainly not anytime soon. In fact, our cameras (eyes) produce rather bad images with just a small properly sharp area (corresponding to the fovia), yet the brain constructs a perceived visual world that is crystal clear – if with the occasional hallucination…

What is really silly is to think that we would stop at the human limitation when we want and need the technology to be better than humans.

We are supposed to eliminate the problems, not replicate them.

Usually, Musk does what he says, and always improves existing techs ๐Ÿ™‚

There are at least two important points here: 1. Currently, Tesla is using video cameras and optical shape/object recognition software to control AutoSteer, in at least some cases, rather than the front-facing radar which is installed in the car. That was made clear when Tesla said that, in the single fatal accident reported when using AutoSteer, the system failed to distinguish between the side of an 18-wheeler rig painted white, and the “brightly lit sky”. Neither radar nor lidar would have had any problem distinguishing between the two, unless the radar or lidar sensor was pointed directly into the sun and blinded by it. 2. Geometry counts. If you have only front-facing, bumper-mounted radar, as Tesla currently uses on its cars, then the field of view for your sensor system is very restricted. The proper place for a scanning sensor, whether it’s radar or lidar, is on top of the car, spinning around where it can see in all directions, and elevated so it can see over intervening vehicles as much as possible. That’s why Google’s self-driving cars have the rotating lidar sensor mounted on top of the car. Bumper-mounted radar, such as Tesla currently uses, is never going to… Read more »

1. Radar would not be blinded by direct sun. Well designed lidar with proper modulated excitation would not be blinded by the sun, either (ie, sun is DC, signal is AC)

2. Geometry does count, but not as much due to height for radar. With radar, phased array would be electronic scanning rather than rotation of lidar. Scattering can be used to radar’s advantage while lidar may not be so easy due to losses / line of sight.

Ultimately, I think radar is a better approach, but it is not as simple as lidar to make them perform the same. Only if I’m not so lazy and bogged down with other stuff…

Isn’t the Dorkmobile only limited to 25mph? What sort of comparison is that?

As for the modded Lexus, those apparently got involved in quite a number of fender benders. Since there is a driver there, nobody is yelling “Autopilot did it”.

Clever trick. Temporal sampling scans, then smoothing the data to create meaningful 3D data for the computer to make safe driving decisions on, will take much processing overhead.

I suspect updated computers in every Tesla with AutoPilot, will soon be required, just to handle pre-crunching radar data for Autonomy.

New testing standards will need to be put in place by existing body (NHTSA, NCAP) and rate the car for events detections (Front, Side, Rear).

They should also rate the car so call autonomous capacity (Fully autonomous means what in which condition, Driver assist means what in which condition)

They way we could get ride of the marketing fluff and creative descriptions.

I personnally believe, what we see on the on the road are different level of driver assist cars. None are fully autonomous, fully autonous car sould detect conditions that would prevent it to work as designed and should disengage.

I could also see some sort of rudimentary questions added to standard drivers license exams in the future.

It’s not like this would acheive much but if these features become more standard in the future it wouldn’t hurt to make sure drivers have at least a basic competency in when and when not to use it.

That is until all cars have level 4 truly autonomous capability.

franky_b said: “They should also rate the car so call autonomous capacity (Fully autonomous means what in which condition, Driver assist means what in which condition)” I was surprised to learn, from a recent comment here on InsideEVs, that there is already a ratings system being used by the industry for levels of autonomy. Quoting from a TechRepublic article: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Level 0: This one is pretty basic. The driver (human) controls it all: steering, brakes, throttle, power. It’s what you’ve been doing all along. Level 1: This semi-autonomous level means that most functions are still controlled by a driver, but some (like braking) can be done automatically by the car. Level 2: In level 2, at least 2 functions are automated, like cruise control and lane-centering. It means that the “driver is disengaged from physically operating the vehicle by having his or her hands off the steering wheel AND foot off pedal at the same time.” The driver must be still always be ready to take control of the vehicle, however. So, level 2 means two functions automated. Easy to remember, right? Level 3: Drivers are still necessary in level 3 cars, but are able to completely shift “safety-critical functions”… Read more »

The problem is, the marketing fluff makes you believe it is fully autonomous, Elon remarks doesn’t make it easier either.

And my point is, it’s not enough to claim it’s a level 2. Car Manifacturer won’t implement it the same way, with the same technology. Is it a 3 stars or a 5 stars Level 2?

Auto-pilot will do nothing for Global Warming, which is one of Musk’s big bugaboos.

If Musk wants to “save the world” with electric cars, then he needs to stop emphasizing falcon wing doors, auto-pilot, bio-hazard filters and get his ass busy and figure out how to mass produce affordable electric cars for the general public.

“get his ass busy”
With three highly successful and disruptive heavy industries, I guess he doesn’t play couch potato very often… ๐Ÿ˜‰

Musk is on his second marriage, he has sacrificed family time to save our sorry oil addicted asses lol

3rd marriage that ended this year.

I would argue auto pilot will help towards the goal of avoiding global warming. It is safer, meaning fewer accidents, fewer repairs, and less new cars to be built as a result.

Less parts and less new cars will most certainly contribute towards the goal of more sustainability and avoid additional global warming.

Eye candy like the falcon wing doors help sell a car that is often tops $100K. Bjorn Nyland and others who test drove the X, have shown in videos, that it often draws a small crowd of admirers at hotels and other places, and makes them feel like an instant celebrity, when the doors go up and they exit the cars.

I have met several Tesla Model S owners. Not one person I met said the main reason was to “save the Earth.” They all thought it was the coolest and most fun car they ever drove. One guy, an aspiring politician at my local university, said his Tesla often attracted a small crowd of admirers, and that he got lots of “friends with benefits” dates from the car after test drives.

Don’t knock it. Isn’t that one of the biggest reason guys want own an awesome car? Danny Zuko builds “a real P…y Wagon” in the classic movie Grease, to get Sandy. Sometimes it works!

Thank you for your critique. I hope it did not take too much time away from your work on solving all of the problems that Elon is struggling with. I will sleep better knowing you are on top of everything.

I’ll mostly agree with Musk, but he made two errors in what he said that need to be pointed out:

1. Lidar is or can be infrared, and thus see though fog.

2. Radar at short waves (microwaves) that would be likely used for such applications has rain attenuation as well.

Again, as said before, active systems, not passive visual, is the key. Your head uses lots of CPU power instead of active ranging to achieve accuracy. It is NOT a perfect system, since it has (your vision) issues with glare and low light situations. That is why bats use active ranging in the dark. And we have gained a new appreciation for how accurate bats can be (for example they can fly through spaced bars).

The bottom line is active systems are more accurate and require less CPU power, which is what we need right now. If you hold out for a perfect artificial intelegence solution, be prepared to wait decades or centuries.

By the way, I work for a company that has achieved very impressive results with stereo cameras and depth perception. I don’t think they would agree with my comments.

But there is a big difference between recognizing objects in front of your computer and driving a car. Big difference.

So, based on Elon’s tweets, the photo of the LIDAR unit on a Model S now makes sense:

They are likely compairing real LIDAR data against their software-sampling / Radar technique as they drive it around.

That photo isn’t that old, so this faux-DAR is all new R&D they’re involved in. The initial results must be somewhat promising, or Elon wouldn’t mention it.